"The sky is... blue": this is what happens (and what could happen) when asking artificial intelligence for a poem

Franz Kafka wrote a short story titled

in which a monkey gains human intelligence and gives a lecture about its past as a wild animal.

Something similar happens when the undersigned here asks ChatGPT about his own limitations: "Are you capable of writing a good poem?"

The language model —as these artificial intelligence tools are called—, which does not flinch at anything, answers: “Although it is possible to create poetry without a human author, most of the poetry considered relevant and significant is usually created by a poet with a distinctive identity and voice.

He admits, with impossible tranquility, that the work of an author who lacks a biography does not usually arouse much interest in readers.

More information

"It works very well, but it's not magic": this is ChatGPT, the new artificial intelligence that exceeds limits

ChatGPT may have become the most prolific poet in the history of literature since it launched just a few months ago.

Thanks to huge

neural networks , they are able to write texts in a few seconds, based on very simple instructions and without plagiarizing a single sentence from the Internet.

However, he cannot be considered the legal author of any of his works.

"Perhaps a machine can create art, but it will not be protected by intellectual property, because for this to happen there must be an author, and as such, except in rare cases, only natural persons can be," says Mario Sol Muntañola, a lawyer. intellectual property expert.

Guillermo Marco and Julio Gonzalo, UNED researchers in the field of Natural Language Processing, have spent years investigating the limits and capabilities of language models such as ChatGPT, and they declare themselves skeptical of "the hype" surrounding this

.

"These models work like networks of artificial neurons, which are like our brain, and what they do is learn to write through a cognitive simulation of how we read," explains Marco, also author of the collection of poems Other Clouds (Rialp, 2019

. , for which he received a Second Prize for the Adonáis Poetry Prize in 2019.

In one of their first experiments, Marco and Gonzalo asked a group of participants to rate six aspects of the synopses of books and movies that this artificial intelligence produced.

"We gave it an invented name and the machine devised an argument for that title," explains Gonzalo.

The result was that the language model scored better than the humans in all the sections except creativity.

The researchers decided to focus on this specific aspect and changed their object of study.

"We started experimenting with poems, because synopses, at least for humans, are not intrinsically creative texts," they say.

Before measuring the results, they proposed to extract a definition of what users understood by creativity.

“For each person it is a different thing,

French Philippe Soupault, considered one of the fathers of automatic poetry.Marc GANTIER (Gamma-Rapho via Getty Images)

The duo concluded that these language models are not designed to be particularly creative.

“Before they are made to imitate than to be original.

They learn to say the least surprising”, they affirm.

Marco illustrates it with an example: "If you give him the sequence: 'The sky is...', and you ask him to complete the sentence, of the 50,000 words he has learned, he will always choose the blue word, because it is the most likely".

ChatGPT's priority, they explain, is to preserve the meaning of the text, which makes it difficult for a sentence to be aesthetically striking.

"Then it turns out that there are people who ask the machine questions that already have such a high originality component that they have no choice but to improvise," Julio Gonzalo specifies.

“I read an example where someone asked ChatGPT to write a Bible-style story about a person who has stuck a butter sandwich in the hole of a VHS video player.

The result was hilarious."

It is also important to consider that ChatGPT does not understand the words it learns.

“All your knowledge is intuitive.

Based on reading they learn what language is.

But they do not have the capacity for reflection or rational thought.

That is precisely why they learn to imitate sonnets, but they are not capable of explicitly realizing that there is a rule that relates a certain type of rhyme to a certain number of verses”, comments Gonzalo.

Marco emphasizes that these language models will always be limited to the input sequence, that is, to the instruction given by a human being.

“They will never have an idea for work.

They will never have intention.

That intention, for the moment, will always come from a human.

What would be truly artistic is for him to refuse to write, or for him to decide to write it in his own way."

surreal neurons

In the book

(Black Box, 2015) the writer Kenneth Goldsmith writes: “Perhaps the great authors of the future will be those who can write the best programs to manipulate, analyze and distribute the practices of language”.

He refers to the possibility that literature becomes a collaboration between machines and humans that does not try to replace human creativity but to increase it.

The writer Jorge Carrión has just published

(Black Box), a book that he has written with a GPT-2 and 3, with the help of the engineers at Taller Estampa.

The starting point of it are

, by André Breton and Philippe Soupault, considered the birth of automatic poetry, in 1919. In his book he compares the arrival and influence of the surrealist group on literature and culture in the first decades of the 20th century, with the expansion of the language models: “If the transition between conscious and unconscious writing characterized those years, writing produced by machine learning and other forms of artificial intelligence is giving ours a particular vibration.”

In a conversation with ICON, he assures that “algorithms write very well, they write almost perfectly, they access areas that are off-limits to humans, but they are still incapable of brilliance, of metaphor, of knowledge of the best poetry.

That does not mean that they do not write better since many poets from experience and their evolution, the poet

”.

However, in the introductory chapter of his book, he does not hesitate to state that the arrival of a technology capable of writing good literature is only a matter of time: "The intelligence of algorithms, robots, neural networks, or programs or formulas that artificial intelligence does not yet exist, and that therefore lacks a name, will end up being capable of metaphor and irony, in new, filmy and undoubtedly literary forms.

You can follow ICON on